Analyzing　network　and　user　behavior　is　very　important　for　a　trusted　network.　We　proposed　a　trust　management　architecture　in　this　study　that　can　evolve　and　maintain　the　behavior　trust　based　on　user＇s　historical　behavior　as　well　as　correspondent　trust　level.　By　collecting　network　traffic　data　from　network　and　analyzing　them　by　means　of　sliding　window　technology,　normal　patterns　of　network　behavior　are　constructed.　Taking　them　as　expected　behavior,　abnormal　data　flow　during　its　runtime　can　be　detected　by　monitoring　the　running　network.　Moreover,　assessment,　prediction　and　control　approaches　of　user＇s　trusted　behavior　based　on　Bayesian　network　are　proposed.　The　relationship　between　quantifiable　evidence　and　the　level　of　trust　assessment　including　multi-attributes　such　as　reliability　is　built.　An　example　is　used　to　illustrate　how　our　model　evolves　and　manages　trust　for　different　trust　attributes.　Experimental　result　shows　that　the　framework　and　approaches　in　this　study　can　control　attacks　in　a　limited　range　and　time　and　predict　trusted　behavior　level,　which　can　improve　the　security　and　reliability　of　servers.　©　2012　Binary　Information　Press.